IMPACT ASSESSMENT FOR AUTOMATEDDRIVING · Simulation of reconstructed case with ADAS / AD ......
Transcript of IMPACT ASSESSMENT FOR AUTOMATEDDRIVING · Simulation of reconstructed case with ADAS / AD ......
EG-342 | 14th November 2018Dr. Felix Fahrenkrog
IMPACT ASSESSMENT FOR AUTOMATED DRIVINGSIP-ADUS WORKSHOP 2018
Tools
TRAFFIC, ACCIDENTS AND TRAFFIC SAFETY.METHODS FOR SAFETY EVALUATION.
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Page 2
Field TestTest Track / Simulator Simulation
Accidentswithout System
Accidentswith System
ReducedRisks
AddedRisks
Multi-dimensional Situation Space
Traffic Situations
Tools
SafetyEvaluation
IMPACT ASSESSMENT. METHODOLOGY - ACCIDENT- VS. TRAFFIC-BASED APPROACH.
Page 3
1) Accident-based
Accident reconstruction
Simulation of reconstructed case with ADAS / AD
Identification of relevant traffic scenarios
ΔSingle-Case
2
Initial constellation
Relevant traffic
scenarios
1
3
1
ΣΔpos- ΣΔneg
3
4Stochastic simulation of traffic situation with ADAS / AD
Stochastic simulation of traffic situation without ADAS / AD
2
2) Traffic-based
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW
FOT & DataAssessment Tools & Data
Impact Assess-ment Method
Impact Assess-ment Tool
TRAFFIC, ACCIDENTS AND TRAFFIC SAFETY.RESEARCH & HARMONIZATION EFFORTS IN SAFETY EVALUATION.
Harmonization Activity
Analyse test tools for (the validation of) automated driving functions.
Provide a common set of relevant situations (database).
Objective
Harmonize / standardize methods for prospective safety performance assessment by virtual simulation of ADAS & AD in order to overcome the issue of variety.
Test of automated driving function on public roads.
Collect data with the automated driving (AD) function.
Provide a transparent software platform that enables the simulation of traffic situations to predict the real-world effectiveness of ADAS and AD.
ResearchActivities
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Page 4
IMPACT ASSESSMENT.METHODOLOGY - P.E.A.R.S.
Representative assessment of active safety and automated driving requires harmonized methods.
For simulation: methods, processes, and models for prospective assessment have to be harmonized.
P.E.A.R.S. is an open working platform to create of a worldwide standard for the evaluation of systems within the pre-crash phase.
WG A “Method, Models and Effectiveness Calculation” WG B “Round Robin Simulation” WG C “Data and Validation & Verification” WG D “ISO and External Communication”
ISO Technical Report 21934-1: ”Road vehicles — Prospective safety performance assessment of pre-crash technology by virtual simulation -- Part 1: State-of-the-art and general method overview“
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Seite 5
Page 6
IMPACT ASSESSMENT.VIRTUAL ASSESSMENT.
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW
Scenario Modelbrakelight (i-1) and visiblebraking?
advance reaction sequence of icar (elapsed ++)
elapsed car i > tbrems?
reaction state icar active
advance position of car using previous speed
YES
YES
rearend collision?
determine acceleration wish of icar
acceleration wish < coasting deceleration?
brakelight icar active
NO
YES
hier quasi entschieden ob „reaction_flag set“?? ja
vehicle data
update speed of icar taking constraints into account
Save accident dataYES
define speed of icar = speed of icar -1
reaction state icar inactive NO
hier quasi entschieden ob „reaction_flag set“??
NO
acute_warning_state?pre_warning ? NO
YES
NO
YESVehicle andFunction Model
determine ttb
iBrake/FCW CCM Warnung
Current warning state
warning state already in effect?
warning state = true, no need to
compute
YES
visibilitydist > distance to object?NO
v > v_object?
YES
B_driver: assumed deceleration of driver
v_egov_objectacc_ego
acc_objectdist_to_object
ttb < threshhold?
Warning alarm = false
NO
Warning alarm = true
YES
NO
NO
iBrake parameter:
v_decrease_iBrake_max zielbremslimit ramptimevsystem prewarningswitch acutewarningswitch zielbremsungswichibswitch1 ibswitch2 ibswitch3 driverBraking_predriverBraking_acute ttbtrigger acc1_ibrake acc2_ibrakeacc_agb wait1 duration1 ramplimitkmh agblimtkmh
compute object stop time
test for crash at ttb_max?
binary search between ttb_min &
ttb_max
YES
compute hypothetical
trajectory
ttb located?
NO
Driver BehaviourModel
Determine minimum buffer distance to
preceeding vehicle
Determine panic distance to preceeding
vehicle
Determine panic distance to preceeding
vehicle
Initialization
Preceeding vehicle visible?
Acceleration wish = 0 NO
Compute perceived distance
(derive from statistical model)
YES
Generate statistical model of perceived
distance
Reaction flag set?
Reaction flag (acute) ?
vi > vdown?
YES
Define acc1 by kinematic formula with perceived_distance:acc1 = -(v-vdown)2 / perceived_distance
Caution: This deceleration can be very strong!!
YES
too close? NO
acc1 = a_coasting (vom Gas gehen)
YES
Normal following: acc1 = min (comfort,
(v-vdown)2 / perceived_distance)
NO
distance to vehicle (i-1) < acceleration perception distance?
Wo ist acceleration perception distance
definiert?
YES
acc2 = minimum (acc1, ai-1)
acc2 = acc1
NO
anticipation flag set? NO
YES
v < v down && too close
v > v down && too close NO
acc1 = coasting
v > v down && not (too close) NO
acc1 = min(-coasting, -(v-vdown)2 / perceived_distance)
acc1 = max(-comfort, -(v-vdown)2 / perceived_distance)
NO
anticipation flag set?
acc2 = acc1 acc2 = min (a_anticipation, acc1)
NO YES
low-pass filter (acc2)
acceleration wish = max(-brakemax and low-pass filter (acc2))
compute anticipation
acceleration from anticipation model
?xi-1< minimum buffer distance?
Draw random numer: probability = samplerate*tick
YES
resample this step?
Probability = probability für Vorausschau (anticipation)?genaue Parametrierung s. Bericht
do over visible downstream cars
YES
NOprevious anticipation
visibility distance
Determine visibility of all preceeding vehicles
vehicle dataperception factor
estimate perceived fictive distance and ?v to each visible downstream
vehicle
minimum acceleration among visible downstream cars
anticipation acceleration
anticipation
anticipation = 0
NO
IMPACT ASSESSMENT.VIRTUAL ASSESSMENT - SIMULATION FRAMEWORK OPENPASS.
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Page 7
OpenPASS is a new software framework for simulation and evaluation of ADAS and automated driving
Join initiative of OEMs (Daimler, VW, Toyota and BMW), Suppliers (Bosch) and other partners (TÜV Süd, itk) with scope of harmonization of simulation tools
Realistic traffic models and simulation investigate interaction between different traffic participants
Fast and efficient simulation consider a high number of situations
Open source approach generate trust and acceptance by authorities and public (Eclipse project: sim@OpenPASS)
BMW VisualizationBMW Visualization
IMPACT ASSESSMENT. IDENTIFICATION OF TOP-SCENARIOS FOR AUTOMATED DRIVING.
Accident data (e.g. GIDAS) / Critical situations (FOT)
Based on specification of function Virtual FOT
Top Scenarios
Simulation of traffic scenarios
Relevant Scenarios
Page 8
A B C
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW
RESULTS FROM EU RESEARCH PROJECT ADAPTIVE.OBSTACLE IN THE LANE.
Page 9
Obstacle in the lane
Scenario Conditions
Probability of remaining crash-free [-]
Parameter ValueSCMDriver
ADF Delta (absolute)
Delta (relative)
Overall - 29.9% 58.2% -28.3% -48,6%
Traffic volume
900veh./h 30.9% 61.6% -30.5% -49,8%
1200 veh./h 28.9% 54.8% -25.9% -47,3%
The survival function
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW
Simulation Time [s]
Pro
porti
on o
f sim
ulat
ion
w/o
col
lisio
n [s
]
RESULTS FROM RESEARCH PROJECT KO-HAF.“MINIMUM RISK MANOEUVRE”.
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Seite 10
Analyse the consequences of two artificial Minimum Risk Manoeuvre (MRM) that consider only braking in the lane (moderate braking vs. strong braking).
The effect of the MRM was simulated for 13 conditional variations in partial factorial design considering the following parameters: speed limit, traffic density and penetration rate of cooperative automated vehicles.
RESULTS FROM RESEARCH PROJECT KO-HAF.“MINIMUM RISK MANOEUVRE”.
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Seite 11
A lower deceleration while the MRM seems to be beneficial in terms of traffic safety benefits, however it requires a longer operation of the system.
V2V communication shows a benefits in the simulation, however larger effects are observed at high penetration rates (>75 %).
0
0.0004
0.0008
0.0012
0 20 40 60 80 100
Col
lisio
n R
ate
[1/k
m]
Penetration rate of passenger cars with cooperative HAD system [%]
Baseline w/o HAD & MRM MRM w. Moderate DecelerationMRM w. Strong Deceleration
*: The case with 100% penetration rate of the passenger cars with a cooperative HAD represents a pure theoretic case that is not realistic.
*
*
OUTLOOK: SAFETY IMPACT ASSESSMENT IN L3PILOT:
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Seite 12
Function description
Identify driving scenarios with potentially reduced
risks (accident scenarios)
Identify driving scenarios with potentially added risks
Definition of to be analysed(relevant) driving scenario
Simulation of relevant driving scenario
Safety effect per driving Scenario (frequency of
accident, collision velocity / position)
Simulation of relevant traffic scenario
Definition traffic scenarios (country specific)
Traffic & infrastructure data Accident data
Change of frequency of driving scenarios at
different penetration rates
Determine accident severity (Injuries) per
driving scenario
Safety impact per driving scenario and penetration
rate
Scen
ario
D
efin
ition
Det
erm
ine
Effe
ctIn
put
Dat
aU
p-sc
alin
g
Input from L3Pilot in-depth evaluation
Identify ODD covered driving
scenario (country specific)
ERiC Method
Weighting factors per driving scenario / country
Incl. Parameters for scenarios, input to modells etc.
The safety impact assessment in L3Pilot aims at:
- Bring together the knowledge and method(s) from different projects and partners.
- Comprehensive assessment of driving scenarios with expected positive and negative effects.
- Consider information about personal attitude towards automated driving from annual survey (e.g. usage).
- Assess the effect for a larger region (aiming at Europe).
- Consider data from the real world pilot tests from different regions in Europe.
PROCESS AND ROLES.VISION OF ACTIVE SAFETY EVALUATION.
Page 13
Neutral (scientific) institutions
Accidentdata
Trafficdata
Humanfactor data
Stochastic scenarios
Industry
Simulationand analysis Results
AD / ADASmodel
Test institute
Weightingof simulated
resultsResults
Caseselection
Spottesting
Rating
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW
SIP-adus Workshop 2018 | 14th November 2018 | Fahrenkrog, BMW Seite 14
THANK YOU